A Comparative Study of Artificial Intelligence Translation Engines based on Multi-modal Natural Language Processing

This study conducts a comparative analysis of five popular artificial intelligence translation engines (Baidu, Doubao, ERNIE Bot, DeepSeek, and ChatGPT). It first introduces three working principles of translation engines, namely rule-based principle, statistical machine principle, and neural machine principle. Then it tests translation performances of the five engines through experiments at word, sentence, and discourse levels, with BLEU as the evaluation indicator. The results show that in word translation, ERNIE Bot has a high BLEU score in general, ChatGPT performs well in technical terms, ERNIE Bot is accurate in political terms, and Doubao and ERNIE Bot are relatively good at explaining cultural terms. In sentence translation, Doubao has the highest overall BLEU score. It especially performs well in subjectless, culture-loaded and redundant sentences, and DeepSeek is also superior in culture-loaded sentences. However, all engines have difficulty with ancient Chinese sentences. In discourse translation, for modern prose, Doubao and DeepSeek have the highest BLEU score, and for ancient poetry, Doubao’s translation captures the rhythm, tone, and elegance of ancient Chinese poetry better. This study provides a useful reference for understanding the advantages and disadvantages of different translation engines, which is conducive to the improvement of translation technology and users’ rational selection of translation tools.